AI art getting better and bigger

One of the highlights of the PastMaster week is seeing the AI art the other guys have mocked up to illustrate the next episode.

If you’ve followed our work on Instagram at all you’ll be familiar with the style. So far we’ve been using Midjourney, which has a nice aesthetic, and latterly we’ve been using a feature called InSwapper to put our own faces into the picture. This has caused great amusement all round.

But my two favourites so far are ‘Machine Gun Wonka’, an imagining of Willy Wonka as a machine gun-toting wild eyed hoodlum, very much riffing on Gene Wilder’s silver screen appearance as the famous chocolatier, and ‘Sack Bolt’, a joyous and astonishingly daft picture of Usain Bolt tripping up in a sack onesie at the Olympics.

Midjourney has been smashing its main rivals in the AI image game so far. But DALL-E, which is owned by OpenAI, has upped its game. It is now a built-in part of ChatGPT 4, the premium version of ChatGPT, which is what we have used for all of our PastMaster adventures so far. Now when you’re entering text prompts to GPT 4, you can include a request for a picture.

It’s much more sophisticated that Midjourney insofar as what you ask for in an image you will more or less get. It’s much better at translating conversational instructions to create an image than Midjourney is. In fact the bonkers stuff Midjourney creates is half the fun.

But one of the best features of the latest iteration of DALL-E is that you can ask it to enhance a particular image over and over again.

This is one of my favourites so far. It’s a picture of jolly old Santa Claus looking happy as he rides his slay across the sky. The poster has asked DALL-E to make Santa Claus ‘more serious’ in each subsequent iteration. At first we get an initial shift as Santa turns from carefree magical present giver to a more austere, dignified character. One presumes presents might be given only to good children perhaps.

Next we get a conservative, very much secular Santa dressed in a plush red smoking jacket, and a statesmanlike confidence. He could easily be a Republican candidate for the White House.

As we seek even more seriousness we see Santa morph from what looks like a corporate head of a toy company, dressed all in black, to a Russian Second World War Commissar, getting ready to order thousands to their graves for the motherland.

Next there is a sudden jump, from menacing Russian psychopath to futuristic soldier armed with a very heavy duty assault rifle. Next Santa keeps the rifle but abandons his throne to go out on patrol, stalking enemies at the head of his death patrol, flanked by a team of robot reindeer. The last of the gun-toting Santas sees the great man as a musclebound cyborg sitting astride a titanic snowmobile.

We then get the shift to the final phase of the metamorphoses from ultimate warrior to god figure, cold and terrible as he drives his team of reindeer forth, bringing frozen wastelands er he travels.

If we trace the trajectory of ‘seriousness’ it looks like this: Kind but stern, to political and probably right wing, to corporate, to ruthless, to cold-hearted, to armed with automatic weapons, to stalking enemies, to godlike. It’s an interesting spectrum of seriousness and I wonder how DALL-E arrived at it.

I tried my own version, urging DALL-E on to draw me a dog of ever increasing size. We started with a St Bernard in someone’s back garden, moved on eventually to a dog that dwarfed skyscrapers and ended up with a dog that appeared to traverse galaxies in outer space.

My favourite bit of the exercise was the accompanying text. Sample: “Here’s the image of an unimaginably massive dog, so large it dwarfs entire cities and landscapes.” It sounded very much like a desperate and obsequious design agency trying to appease a mad and unreasonable client. “Sir, that dog really is already almost too big to comprehend. I really don’t think you need a bigger one.” And yet it complies, seeking out some way to assuage to insatiable hunger of the big dog craving idiot.

What I’ve found is that whatever superlative you ask it to apply to a picture: happiest, biggest, most serious, skinniest, the further you go, the more likely it is to end in outer space. It seems that somehow space really is the final frontier when it comes to AI art.